1. Field of the Invention
The invention relates to an optical device with hybrid integrated optical waveguide chips. In particular, the invention relates to an optical device using flip-chip method to hybrid integrate one or more optical chips on an optical bench and both the chips and the optical bench include multiple optical waveguides with different distances to compensate for alignment errors.
2. Description of the Related Art
Optical interconnects are adopted in data communications at unprecedented rate as more bandwidth and longer transmission reach are required by mega datacenters for applications from social networks, cloud service, to big data analysis and high performance computing. Unlike optical transceiver modules or subsystems made of ultrahigh performance discrete components in telecommunications, lower cost, more compact and more power efficient optical transceivers or engines are demanded in data communications. Integrating multiple optical components or chips such as lasers, modulators, photodetectors, switches, attenuators and etc. on an optical bench chip to form a hybrid integrated optical device is one way to reduce assembling cost and footprint.
In such hybrid integrated optical devices, passively placing and bonding the optical chips on optical benches is highly preferred as it enables automated low cost assembling for massive volume production required by huge data communications market. However, unlike the mature integrated circuit (IC) fully automated packaging processes, assembling these optical chips requires very precise alignment in the range of micrometers or less because these chips and optical benches usually include tiny optical waveguides which must be well aligned with each other to form an optical transmission path.
Borrowing from the IC packaging industry, people have been trying to use the tools called flip-chip bonder to bond the optical chips upside down onto an optical bench. Because the optical waveguides are almost always formed on the top side of an optical chip or an optical bench by some semiconductor or other wafer processing techniques, the distance between the optical waveguide and the top surface is well controlled. By placing an optical chip upside down onto an optical bench and with some pre-defined spacer structures on the optical bench, the optical waveguide alignment in the direction perpendicular to the surface (out-of-plane) of the optical chip and the optical bench can be precisely controlled. This flip-chip bonding approach has been widely discussed.
On the other hand, the alignment in the directions parallel to the surface (in-plane) can only be controlled by the flip-chip bonder's accuracy and bonding process control. A modern top-of-the-line flip-chip bonder can achieve a +/−0.5 micrometer alignment accuracy, however, in practice, the bonding involving processes such as thin metal solder melting, adhesive curing and etc. inevitably contributing to final alignment error due to physical movement of the chip under temperature, stress and/or phase changes. The final alignment error (3σ confidence interval) is usually +/−2 micrometers or worse from the statistics of our experiment. The alignment in in-plane waveguide propagation direction is relatively tolerant and satisfied with this alignment error while the in-plane direction perpendicular to waveguide propagation requires accurate alignment, especially for small optical waveguides such as those in lasers. To increase the alignment tolerance in this direction, people tried to include either a taper structure at the end of the waveguide or a lens structure in order to expand the optical beam for more tolerant alignment. However, including a taper structure as part of the optical waveguide requires design change of the optical chips which prohibits the use of widely available and proven commercial chips as well as, in many cases, harms device performance. The lens which can be used in such condition cannot be made monolithically on the optical bench and has to be installed separately which introduces additional alignment error during the assembling. These and similar methods have been proposed but none of them is being adopted in mass production due to above-stated issues.